© 2017, Springer-Verlag GmbH Germany. Very little is known about the effects of specific practice on motor learning of predictive balance control during novel bipedal gait. This information could provide an insight into how the direction and magnitude of predictive errors during acquisition of a novel gait task influence transfer of balance control, as well as yield a practice protocol for the restoration of balance for those with locomotor impairments. This study examined the effect of a variable practice paradigm on transfer of a novel asymmetric gait pattern in able-bodied individuals. Using a split-belt treadmill, one limb was driven at a constant velocity (constant limb) and the other underwent specific changes in velocity (variable limb) during practice according to one of three prescribed practice paradigms: serial, where the variable limb velocity increased linearly; random blocked, where variable limb underwent random belt velocity changes every 20 strides; and random practice, where the variable limb underwent random step-to-step changes in velocity. Random practice showed the highest balance control variability during acquisition compared to serial and random blocked practice which demonstrated the best transfer of balance control on one transfer test. Both random and random blocked practices showed significantly less balance control variability during a second transfer test compared to serial practice. These results indicate that random blocked practice may be best for generalizability of balance control while learning a novel gait, perhaps, indicating that individuals who underwent this practice paradigm were able to find the most optimal balance control solution during practice.